From: Sage Weil <sage@redhat.com>
Date: Tue, 2 Dec 2014 02:15:59 +0000 (-0800)
Subject: osd: tolerate sessionless con in fast dispatch path
X-Git-Tag: v0.91~165^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fpull%2F3048%2Fhead;p=ceph.git

osd: tolerate sessionless con in fast dispatch path

We can now get a session cleared from a Connection at any time.  Change
the assert to an if in ms_fast_dispatch to cope.  It's pretty rare, but it
can happen, especially with delay injection.  In particular, a racing
thread can call mark_down() on us.

Fixes: #10209
Backport: giant
Signed-off-by: Sage Weil <sage@redhat.com>
---

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 283a5178c5e..f1c28457cb6 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -2648,7 +2648,8 @@ PG *OSD::get_pg_or_queue_for_pg(const spg_t& pgid, OpRequestRef& op)
 {
   Session *session = static_cast<Session*>(
     op->get_req()->get_connection()->get_priv());
-  assert(session);
+  if (!session)
+    return NULL;
   // get_pg_or_queue_for_pg is only called from the fast_dispatch path where
   // the session_dispatch_lock must already be held.
   assert(session->session_dispatch_lock.is_locked());
@@ -5407,14 +5408,15 @@ void OSD::ms_fast_dispatch(Message *m)
   }
   OSDMapRef nextmap = service.get_nextmap_reserved();
   Session *session = static_cast<Session*>(m->get_connection()->get_priv());
-  assert(session);
-  {
-    Mutex::Locker l(session->session_dispatch_lock);
-    update_waiting_for_pg(session, nextmap);
-    session->waiting_on_map.push_back(op);
-    dispatch_session_waiting(session, nextmap);
+  if (session) {
+    {
+      Mutex::Locker l(session->session_dispatch_lock);
+      update_waiting_for_pg(session, nextmap);
+      session->waiting_on_map.push_back(op);
+      dispatch_session_waiting(session, nextmap);
+    }
+    session->put();
   }
-  session->put();
   service.release_map(nextmap);
 }
 
@@ -5424,10 +5426,12 @@ void OSD::ms_fast_preprocess(Message *m)
     if (m->get_type() == CEPH_MSG_OSD_MAP) {
       MOSDMap *mm = static_cast<MOSDMap*>(m);
       Session *s = static_cast<Session*>(m->get_connection()->get_priv());
-      s->received_map_lock.Lock();
-      s->received_map_epoch = mm->get_last();
-      s->received_map_lock.Unlock();
-      s->put();
+      if (s) {
+	s->received_map_lock.Lock();
+	s->received_map_epoch = mm->get_last();
+	s->received_map_lock.Unlock();
+	s->put();
+      }
     }
   }
 }
@@ -5632,13 +5636,15 @@ bool OSD::dispatch_op_fast(OpRequestRef& op, OSDMapRef& osdmap)
   if (msg_epoch > osdmap->get_epoch()) {
     Session *s = static_cast<Session*>(op->get_req()->
 				       get_connection()->get_priv());
-    s->received_map_lock.Lock();
-    epoch_t received_epoch = s->received_map_epoch;
-    s->received_map_lock.Unlock();
-    if (received_epoch < msg_epoch) {
-      osdmap_subscribe(msg_epoch, false);
+    if (s) {
+      s->received_map_lock.Lock();
+      epoch_t received_epoch = s->received_map_epoch;
+      s->received_map_lock.Unlock();
+      if (received_epoch < msg_epoch) {
+	osdmap_subscribe(msg_epoch, false);
+      }
+      s->put();
     }
-    s->put();
     return false;
   }
 
@@ -8098,7 +8104,7 @@ void OSD::handle_replica_op(OpRequestRef& op, OSDMapRef& osdmap)
     op->send_map_update = should_share_map;
     op->sent_epoch = m->map_epoch;
     enqueue_op(pg, op);
-  } else if (should_share_map) {
+  } else if (should_share_map && m->get_connection()->is_connected()) {
     C_SendMap *send_map = new C_SendMap(this, m->get_source(),
 					m->get_connection(),
                                         osdmap, m->map_epoch);